Question about big Gig E rings

SONET is the way to go if you want extremely fast failover detection and recovery. SONET is will do this in micro-seconds, Ethernet would do this in less than a minute. Since you have "North" and "South" routes built out, it looks to me that SONET was the choice when the fiber routes were engineered. If the plan was using Ethernet, the fiber build-out plan would be different. For Ethernet, you would have different paths and fewer fibers would be required, but it would not be a resilient and fail-over would not be anywhere near as fast. SONET provides virtually no packet loss due to a fiber break or equipment failure.

Scott

The North/South routes were more to keep the network from getting cut in half in case someone decides to DUI and takes out a poll, and then can't get his medical records because the hospital is offline.

Extra fibers were put in so we wouldn't need to go back if we needed more bandwidth than was available on a single pair.

"thcollicutt" ha scritto nel messaggio news: snipped-for-privacy@q66g2000hsg.googlegroups.com...

Hi,

SONET is good but legacy; if you do not need to reuse a SONET network, you have better alternatives ( well, in fact you could have either SONET and other technologies at the same time if you need to ) Look at the Multi Service Equipment from Cisco ( ONS 15xxx ). You coulhave have SONET,RPR,Ethernet,old PDH ( T1/E1,etc. ) and also CWMA and DWMA.

Regards, Gabriele

you can do fast rerouting with Ethernet pipes - as long as your route at the nodal points.

We have shown OSPF based rerouting on Cisco Cat 6500s at sub 50 mSec in the lab - but it takes a fair amount of tuning to get there.

you need the GigE links to give "loss of light" on a fault so the equipment can react immediately - shouldnt be a problem if you are driving dark fibre directly.

we are just commissioning a similar network for a customer - dual central hubs with GigE to each location.

However we are using "legacy" modern SDH to drive the plumbing at STM-64 /

9.8 Gbps and carving the GigEs out of that since we need non IP services as well.

MPLS is probably the best way to do this, if all you need to do is move packets.

Fast reroute on MPLS equipment converges as quickly as SDH (european flavour of SONET used here in UK).

Although MPLS can support lots of VPN structures "on top", the boring conventional service is routed IP VPNs - which means these are the ones which are well known, well behaved and that a lot of trained engineers know how to build and look after.

I have worked on Cisco MPLS a lot - good, loads of features, but expensive and you need to lab test to find out which code versions will work with the features you need. Just about all the other router manufacturers also support it, as well as traditional telecoms suppliers (Alcatel specifically seems to make reliable MPLS stuff)

What you have said implies that routed IP is going to do what you need - so start with that as the design baseline.

MPLS allows each of your logical overlays to choose their own addressing, QoS, topology and various other things, so will reduce the complexities of glueing several networks together and making them play nice.

SONET / SDH equipment has benefited from the same improvements in performance / chip scale as anything else made of complex electronics.

The problem you may have is the kit is almost always optimised for a telco and use on a large scale, so needs expensive management tools and so on, and you may find maintenance etc from a 3rd party "harder" than routing style equipment.

If you go down the SONET route then you are going to be acting like a telco, and be supplying pipes rather than IP networks to each (unless you build that as another layer) - that may be a good design tradeoff depending on what you want.

>

50 ms or less failover is good, but currently we have a network of rented lines with no failover and a 4 hour repair time. I think failover of a minute or two may fail the desired time for VoIP, but it is significantly faster than anything we have now.

I must say that the carriers we have been using have been quite good at getting these lines back up and running on the rare occasion that they go down, so I don't want to give the impression that the lines go down all the time, or take a really long time to repair. I am just noting the difference between SONET and current repair times.

I am not ready to jump into SONET for a new network, and I don't want a 400+ km of fiber being controlled by Spanning tree. So, I am looking at having multiple areas, and routing between them. The server group has suggested that the central ring, which connects all the hub sites, to be layer 2 so they can have the same IP subnet present at each hub site for them to do their server redundancy. This looks like it may complicate the design a bit. My experience with MPLS is knowing what it is.

I'm going to have to go back and reread my stuff on OSPF and Spanning tree. I was just curious, since I have a vendor coming in to explain some optoelectronics they have for sale, whether my wish to remain with ethernet based solutions is a reasonable one, and whether others have been able to do someting similar.

addressing,

In theory from a server perspective this is great, but very problematic for the plumbing when L2 goes between sites.

they really need to think about using a server resilience scheme that they can use at layer 3 or higher.

This

this is serous understatement - since L2 links between sites seem to fail in various ways - most of which i only found when they happened....

The problem is that any such failure where there are devices on the 2 sites subnet will end up with the "split subnet" problem.

this is where there are 2 bits of the same subnet which "should" stay connected dont. This leads to big connectivity holes in a notionally resilient network.

Routers then deliver packets to which ever section of the subnet is closest - because the implicit assumption for the routers is that subnets are "atomic" and dont get carved into bits.

The same can happen with router to router links - but the routers use a routing protocol to detect such transit path problems and will reroute around the subnet.

My experience with

OSPF would be my preferred way to do something like this.

I was just curious, since I have a vendor coming in to explain

optoelectronics they have for sale, whether my wish to remain

Thanks. This pretty much sums up what my thoughts were when I sat down to figure out how to do this.